Fix GC Statistics and time measuring

[ramLlama]

getrusage(RUSAGE_THREAD, ...) is too low precision to get accurate GC times.

[ramLlama]

So, I can calculate the wallclock time taken by GC using clock_gettime(), but I no longer have the user/sys split that rusage has. As such, these timings don't really fit into the GC rusage. Any thoughts on this?

A couple options:

• keep both -- The GC rusage remains, but is imprecise (perhaps to the point of being useless). Add additional MLton.GC.Statistics functions that return time taken.
• fold the result from clock_gettime() into the rusage, perhaps as the user field. This is most transparent since the API remains unchanged, but the rusage data is now wrong, since all GC time will be in the user slot.

[shwestrick]

We also need to update MLton.GC.Statistics for the new algorithms. What is a reasonable interface?

[shwestrick]

Closing this issue, as it is subsumed by #106.